CN112036079A - Diesel engine multi-information fusion diagnosis method based on ANFIS - Google Patents

Diesel engine multi-information fusion diagnosis method based on ANFIS Download PDF

Info

Publication number
CN112036079A
CN112036079A CN202010831743.9A CN202010831743A CN112036079A CN 112036079 A CN112036079 A CN 112036079A CN 202010831743 A CN202010831743 A CN 202010831743A CN 112036079 A CN112036079 A CN 112036079A
Authority
CN
China
Prior art keywords
diesel engine
parameters
anfis
parameter
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010831743.9A
Other languages
Chinese (zh)
Other versions
CN112036079B (en
Inventor
费景洲
韩雨婷
王忠巍
曹云鹏
袁志国
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Engineering University
Original Assignee
Harbin Engineering University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202010831743.9A priority Critical patent/CN112036079B/en
Publication of CN112036079A publication Critical patent/CN112036079A/en
Application granted granted Critical
Publication of CN112036079B publication Critical patent/CN112036079B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/02—Reliability analysis or reliability optimisation; Failure analysis, e.g. worst case scenario performance, failure mode and effects analysis [FMEA]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Combined Controls Of Internal Combustion Engines (AREA)

Abstract

The invention discloses a diesel engine multi-information fusion diagnosis method based on ANFIS. Step 1: outputting a result set F and inputting a parameter set S; fault database U1(ii) a Step 2: carrying out normalization processing on the data set; and step 3: carrying out analytic hierarchy process on the data set to obtain a parameter weight value; and 4, step 4: for generating initial rule structure Q of diesel engine fault diagnosis model0(ii) a And 5: will Q0Obtaining a rule structure Q of a diesel engine fault diagnosis model after training1(ii) a Step 6: to Q1Inputting the actual operation of the diesel engine after normalization processingAnd the row parameter S is recorded as an input parameter set S ', and after fuzzy fitting and rule fitting are carried out on the S', a diesel engine fault diagnosis model based on ANFIS is generated and output. The invention improves the engineering practicability of the ANFIS in the aspect of the multi-information fusion diagnosis technology of the diesel engine.

Description

Diesel engine multi-information fusion diagnosis method based on ANFIS
Technical Field
The invention belongs to the technical field of diesel engine fault diagnosis; in particular to a diesel engine multi-information fusion diagnosis method based on ANFIS.
Background
The diesel engine is the most widely applied machine type in various power machines at present, and the development of the diesel engine technology has important influence on various aspects of industry and agriculture, traffic transportation, national defense construction and the like in China. The diesel engine is easy to generate various faults due to the reasons of complex structure, severe working conditions, variable operating conditions and the like. The failure of the diesel engine can affect the operation safety of equipment and a system, and even cause great loss in personnel and property aspects in case of serious failure, so that the development of the technical research of diesel engine failure diagnosis has great significance for ensuring the safe and efficient operation of the diesel engine.
Among the existing methods for monitoring and diagnosing faults of diesel engines, a performance parameter analysis method, an oil analysis method, a sound vibration signal analysis method, an expert system, a neural network method and a multi-information fusion method are the most common and effective methods. The multi-information fusion diagnosis method can eliminate the self uncertainty of equipment and sensors in the single-source information diagnosis process, improve the precision and reliability of fault diagnosis, and become important research content in the aspect of fault diagnosis of the existing diesel engine. The artificial neural network method has the characteristics of large-scale parallel processing, self-organizing learning, good nonlinear mapping capability and the like, and has wide application prospect in the aspect of multi-information fusion diagnosis of the diesel engine. The existing artificial neural network method has the problems that due to the lack of accurate expression between network parameters and mathematical functions, a model has a 'black box' problem of unclear internal structure, and a BP (back propagation) forward neural network has the problems of local minimum values and the like, so that the development of the neural network method in the aspect of multi-information fusion diagnosis of a diesel engine is limited.
Disclosure of Invention
The invention provides an ANFIS-based multi-information fusion diagnosis method for a diesel engine, which improves the engineering practicability of the ANFIS in the aspect of multi-information fusion diagnosis technology of the diesel engine.
The invention is realized by the following technical scheme:
a diesel engine multi-information fusion diagnosis method based on ANFIS comprises the following steps,
step 1: selecting typical faults of the diesel engine as an output result set F, and selecting kinetic and thermodynamic parameters which are convenient to measure in actual operation of the diesel engine as an input parameter set S; fault database U trained by taking known fault parameter set as model1；
Step 2: the output result set F, the input parameter set S and the fault database U of the step 1 are combined1Normalizing the data set;
and step 3: carrying out analytic hierarchy process on the data set subjected to the normalization process in the step 2 to obtain a parameter weight value;
and 4, step 4: combining the parameter weight values in the step 3 with the ANFIS optimized by the subtractive clustering method to generate an initial rule structure Q of the diesel engine fault diagnosis model0；
And 5: the initial rule structure Q of the step 40Obtaining a rule structure Q of a diesel engine fault diagnosis model after training1；
Step 6: rule structure Q of fault diagnosis model of diesel engine1Inputting characteristic parameters in the diesel engine actual operation parameter set S subjected to normalization processing, recording the characteristic parameters as an input parameter set S ', performing fuzzy fitting and rule fitting on the input parameter set S', and generating a diesel engine fault diagnosis model based on ANFIS.
Further, the fault types in step 1 include: normal operation f0Single cylinder misfire f1Exhaust pipe leakage f2Dirt resistance f of air compressor3Air filter clogging f4Poor lubrication f5；
The input parameters include: effective power s1Fuel consumption s2Front vortex exhaust temperature s3Post-vortex exhaust temperature s4Vortex front exhaust pressure s5Front temperature s of intercooler6And the rear temperature s of the intercooler7Front pressure s of intercooler8And intercooler rear pressure s9Inlet pressure s10After cylinder average temperature s11。
Further, the data set is normalized in step 2, that is, each group of Si in the set S is normalized by the maximum and minimum method
Figure BDA0002638254680000021
Wherein, i is the number of parameters i is 1, …, 11, k is the number of data groups siminK is the mean of the data sequence, 1, 2, … …; simaxIs the variance of the data; setting the normalization range to be 0-1; normalization eliminates the magnitude difference between different types of input parameters.
Further, the step 3 of performing analytic hierarchy process on the data set specifically includes obtaining a difference between the fault data and the normal data, establishing a data feature matrix, obtaining a feature value and a feature vector of the matrix, comparing the maximum feature vector, and performing consistency detection to obtain a parameter weight value.
Further, the analytic hierarchy process specifically includes seeking a parameter variation degree, that is, seeking a difference value between a fault parameter and a normal parameter in a known set, and forming a parameter variation matrix a by using the difference value if a ratio of the difference value between the factor i and the factor j is equal to
Figure BDA0002638254680000025
Then the ratio of the difference between factor j and factor i is
Figure BDA0002638254680000022
Because the square matrix A ═ aij)n×nThe requirements are met,
Figure BDA0002638254680000023
so that the square matrix A is (a)ij)n×nIs a positive and negative matrix;
but the square matrix A ═ aij)n×nIt is not necessarily the case that,
Figure BDA0002638254680000024
the matrix A ═ aij)n×nNot necessarily a consistency matrix;
according to the consistency matrix characteristics:
if A is ═ aij)n×nMaximum eigenvalue λ ofmaxThe corresponding feature vector is w ═ w (w)1，…，wn)TThen, then
Figure BDA0002638254680000031
Wherein w is the parameter weight value;
by analytic hierarchy process basic structure, before obtaining the weighted value, should construct square matrix A and carry out the conformance test, through the conformance proportion CR:
Figure BDA0002638254680000032
when CR is less than 0.10, the matrix is a consistency matrix, otherwise, the matrix is modified;
the numerical value of CI is calculated and obtained according to the square matrix A, the numerical value of RI is obtained by looking up an RI value table of the analytic hierarchy process random consistency, the numerical value of RI is determined by the order n of the square matrix A, the numerical value of RI can also be obtained by calculating the square matrix A, and the numerical value of RI is obtained by adopting a table look-up method in the invention.
And solving the characteristic value and the characteristic vector of the difference value square matrix A, and carrying out analytic hierarchy process to obtain the weight value of the parameter.
Further, the initial rule structure Q of the diesel engine fault diagnosis model in the step 40Specifically, the training steps and the maximum error are set, and a fault database U is used1For the initial structure Q0Carry out trainingTraining, generating rule structure Q of diesel engine fault diagnosis model after training1。
Further, the subtractive clustering optimized ANFIS, wherein s1、s2Is an input parameter, and under the condition of multi-parameter input, the structure generation conditions are as follows:
the first layer is a blurring layer, and the expression is as follows:
Figure BDA0002638254680000033
wherein A isi、BiRepresenting a fuzzy set;
Figure BDA0002638254680000034
and
Figure BDA0002638254680000035
are respectively s1，s2Selecting the expression of the bell-shaped function as follows:
Figure BDA0002638254680000036
wherein the content of the first and second substances,
Figure BDA0002638254680000037
the parameter set is advanced, the value of the parameter set is continuously updated by feedback in the training stage, and finally a conclusion parameter set in the rule is formed;
the second layer is used for realizing the operation of the fuzzy set in the first layer, the output of the first layer is represented as points in the layer, the algebraic product of signals is output through the calculation of the layer, the output result of each point is represented as a rule applicability, and the expression is as follows:
Qi＝μAi(s1)×μBi(s2)，i＝1，2 (8)
the third layer is to normalize the excitation intensity of each rule, the node of the layer is a fixed node, and the output is the applicability of the rule and all rules; the expression is as follows:
Figure BDA0002638254680000041
all nodes of the fourth layer are self-adaptive to calculate the output of each rule; the expression is as follows:
Figure BDA0002638254680000042
the fifth layer is an output layer which is used for calculating the sum of all transmitted signals as an output signal; the expression is as follows:
Figure BDA0002638254680000043
for advanced parameter modification, a suitable set of parameters is found such that,
Figure BDA0002638254680000044
wherein f is the actual output;
Figure BDA0002638254680000045
and outputting the model.
Further, the subtractive clustering method is applied to the first layer and the second layer, the subtractive clustering is a density algorithm, and is used for finding the center of the data, and the density of each data point is firstly calculated to obtain a density index:
Figure BDA0002638254680000046
finding out the data with the maximum density index as a first clustering center, removing the density of the point, and calculating the density indexes of multiple suspicious points;
Figure BDA0002638254680000047
finding the maximum density index, taking the point as a clustering center, and sequentially circulating until the conditions are met:
Figure BDA0002638254680000048
where is a set, small positive number.
Further, the generating of the fault diagnosis preliminary model FIS includes the steps of processing input parameters and judging a rule structure Q0And a complete algorithm structure of the result output step. The FIS generated by training is used for the fault diagnosis preliminary model: applying training data to rule Structure Q0Training, namely training the generated FIS by using the complete fault data of the known result set F and the input parameter set S, namely training the precondition parameters and the conclusion parameters, and improving the diagnosis precision of the model by using the training parameters;
for the precondition parameters, a back propagation algorithm is applied, for the conclusion parameters, a linear least square estimation algorithm is adopted to adjust the parameters, the input signals are firstly transmitted to the fourth layer along the network forward direction in each iteration, at the moment, the precondition parameters are fixed, and the conclusion parameters are adjusted by adopting the least square estimation algorithm until the model simulation runs to the fifth layer output layer;
the obtained error signal is propagated reversely along the diagnostic model ANFIS structure, so as to adjust the precondition parameters; in this way, the optimal values of the parameters are found, and the training is finished when the training reaches the specified number of steps, generating the fault diagnosis structure FIS'.
Furthermore, during diagnosis, diesel engine fault input parameter data S' and training data S are integrated and then are subjected to normalization processing, the processed data are input into a parameter set S ", the input parameter set S" is used as an input parameter and is input into a fault diagnosis ANFIS structure for simulation operation, the obtained result is a fault type, and finally a diesel engine fault diagnosis model which integrates various parameters and generates a judgment result is formed.
The invention has the beneficial effects that:
1. the invention solves the problem of 'calculated amount explosion' in the multi-information fusion process of the conventional ANFIS algorithm, and has the advantages of more fusion parameters, small algorithm calculated amount and the like;
2. the method realizes automatic optimization of algorithm inference rules without depending on expert experience conditions, and solves the problem that the traditional subjective analysis method highly depends on expert subjective experience and the iterative optimization algorithm increases modeling complexity;
3. the model can realize the automatic adjustment of the model structure and the inference rule according to the change of the type and the quantity of the input parameters, and has self-learning capability; along with the continuous accumulation and enrichment of the operation data of the diesel engine, the diagnosis precision of the model can be gradually improved.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
FIG. 2 is a schematic diagram of the ANFIS rule structure of the present invention using subtractive clustering with a hierarchal analysis.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
A diesel engine multi-information fusion diagnosis method based on ANFIS comprises the following steps,
step 1: selecting typical faults of the diesel engine as an output result set F, and selecting kinetic and thermodynamic parameters which are convenient to measure in actual operation of the diesel engine as an input parameter set S; fault database U trained by taking known fault parameter set as model1；
Step 2: the output result set F, the input parameter set S and the fault database U of the step 1 are combined1Normalizing the data set; eliminating the difference between the magnitude and the unit of the parameters and reducing the difficulty of the fuzzification processDifficulty;
and step 3: carrying out analytic hierarchy process on the data set subjected to the normalization process in the step 2 to obtain a parameter weight value;
and 4, step 4: combining the parameter weight values in the step 3 with the ANFIS optimized by the subtractive clustering method to generate an initial rule structure Q of the diesel engine fault diagnosis model0；
And 5: the initial rule structure Q of the step 40Obtaining a rule structure Q of a diesel engine fault diagnosis model after training1；
Step 6: rule structure Q of fault diagnosis model of diesel engine1Inputting characteristic parameters in the diesel engine actual operation parameter set S subjected to normalization processing, recording the characteristic parameters as an input parameter set S ', performing fuzzy fitting and rule fitting on the input parameter set S', and generating a diesel engine fault diagnosis model based on ANFIS.
Further, the fault types in step 1 include: normal operation f0Single cylinder misfire f1Exhaust pipe leakage f2Dirt resistance f of air compressor3Air filter clogging f4Poor lubrication f5；
The input parameters include: effective power s1Fuel consumption s2Front vortex exhaust temperature s3Post-vortex exhaust temperature s4Vortex front exhaust pressure s5Front temperature s of intercooler6And the rear temperature s of the intercooler7Front pressure s of intercooler8And intercooler rear pressure s9Inlet pressure s10After cylinder average temperature s11。
The multi-parameter selection is beneficial to judging the running condition of the diesel engine in an all-around manner, so that the diagnosis result is more accurate, and the fault diagnosis model can comprehensively analyze the multi-parameter information of the diesel engine due to the diversified reference parameters, thereby improving the diagnosis precision and reliability.
Further, the data set is normalized in step 2, that is, each group of Si in the set S is normalized by the maximum and minimum method
Figure BDA0002638254680000061
Wherein, i is the number of parameters i is 1, …, 11, k is the number of data groups siminK is the mean of the data sequence, 1, 2, … …; simaxIs the variance of the data; setting the normalization range to be 0-1; normalization eliminates the magnitude difference between different types of input parameters. The normalized data size of a single parameter does not indicate the diesel engine operating conditions.
Further, the step 3 of performing analytic hierarchy process on the data set specifically includes obtaining a difference between the fault data and the normal data, establishing a data feature matrix, obtaining a feature value and a feature vector of the matrix, comparing the maximum feature vector, and performing consistency detection to obtain a parameter weight value.
Further, the method is to analyze the influence of the intensity of parameter change on the result by applying an analytic hierarchy process. According to the reduction process of the analytic hierarchy process, the parameters with large influence on the result are decomposed into weight, rules and levels, and on the basis of outputting the weight, the parameters are compared to finally set the key parameters. The invention does not make a decision according to the analytic hierarchy process, and only searches the weight value of the corresponding parameter according to the analytic hierarchy process. The analytic hierarchy process includes finding out parameter variation degree, namely finding out the difference between fault parameter and normal parameter in known set, forming parameter variation matrix A with the difference, and determining the ratio of the difference between factor i and factor j
Figure BDA0002638254680000065
Then the ratio of the difference between factor j and factor i is
Figure BDA0002638254680000062
Because the square matrix A ═ aij)n×nThe requirements are met,
Figure BDA0002638254680000063
so that the square matrix A is (a)ij)n×nIs a positive and negative matrix;
but the square matrix A ═ aij)n×nIt is not necessarily the case that,
Figure BDA0002638254680000064
the matrix A ═ aij)n×nNot necessarily a consistency matrix;
according to the consistency matrix characteristics:
if A is ═ aij)n×nMaximum eigenvalue λ ofmaxThe corresponding feature vector is w ═ w (w)1，…，wn)TThen, then
Figure BDA0002638254680000071
Wherein w is the parameter weight value;
by analytic hierarchy process basic structure, before obtaining the weighted value, should construct square matrix A and carry out the conformance test, through the conformance proportion CR:
Figure BDA0002638254680000072
when CR is less than 0.10, the matrix is a consistency matrix, otherwise, the matrix is modified;
the numerical value of CI is calculated and obtained according to the square matrix A, the numerical value of RI is obtained by looking up an RI value table of the analytic hierarchy process random consistency, the numerical value of RI is determined by the order n of the square matrix A, the numerical value of RI can also be obtained by calculating the square matrix A, and the numerical value of RI is obtained by adopting a table look-up method in the invention.
The construction matrix A is specially selected according to the parameter difference, the matrix is not constructed according to the importance of the original analytic hierarchy process, absolute values are taken from the numerical values of the construction matrix A, and all data belong to 0-1, so that the consistency test is only used as a part determined by detection; and solving the characteristic value and the characteristic vector of the difference value square matrix A, and carrying out analytic hierarchy process to obtain the weight value of the parameter.
Further, the initial rule structure Q of the diesel engine fault diagnosis model in the step 40Specifically, the training steps and the maximum error are set, and a fault database U is used1For the initial structure Q0Training is carried out, and a rule structure Q of a diesel engine fault diagnosis model after training is generated1。
Further, the subtractive clustering optimized ANFIS, wherein s1、s2Is an input parameter, and under the condition of multi-parameter input, the structure generation conditions are as follows:
the first layer is a fuzzy layer, which is to divide the input data according to the membership degree for fuzzification, and the expression is as follows:
Figure BDA0002638254680000073
wherein A isi、BiRepresenting a fuzzy set;
Figure BDA0002638254680000074
and
Figure BDA0002638254680000075
are respectively s1，s2Selecting the expression of the bell-shaped function as follows:
Figure BDA0002638254680000076
wherein the content of the first and second substances,
Figure BDA0002638254680000077
the parameter set is advanced, the value of the parameter set is continuously updated by feedback in the training stage, and finally a conclusion parameter set in the rule is formed;
the second layer is used for realizing the operation of the fuzzy set in the first layer, the output of the first layer is represented as points in the layer, the algebraic product of signals is output through the calculation of the layer, the output result of each point is represented as a rule applicability, and the expression is as follows:
Qi＝μAi(s1)×μBi(s2)，i＝1，2 (8)
the third layer is to normalize the excitation intensity of each rule, the node of the layer is a fixed node, and the output is the applicability of the rule and all rules; the expression is as follows:
Figure BDA0002638254680000081
all nodes of the fourth layer are self-adaptive to calculate the output of each rule; the expression is as follows:
Figure BDA0002638254680000082
the fifth layer is an output layer which is used for calculating the sum of all transmitted signals as an output signal; the expression is as follows:
Figure BDA0002638254680000083
for advanced parameter modification, a suitable set of parameters is found such that,
Figure BDA0002638254680000084
wherein f is the actual output;
Figure BDA0002638254680000085
and outputting the model.
Further, the subtractive clustering method is applied to the first layer and the second layer, the subtractive clustering is a density algorithm, and is used for finding the center of the data, and the density of each data point is firstly calculated to obtain a density index:
Figure BDA0002638254680000086
finding out the data with the maximum density index as a first clustering center, removing the density of the point, and calculating the density indexes of multiple suspicious points;
Figure BDA0002638254680000087
finding the maximum density index, taking the point as a clustering center, and sequentially circulating until the conditions are met:
Figure BDA0002638254680000088
where is a set, small positive number.
The subtractive clustering method has the advantages that the algorithm is operated quickly, the complexity and the data dimension are in a linear relation, model data and rule number are reduced by applying the subtractive clustering algorithm, and the rule number overflow phenomenon caused by the ANFIS when the input parameters are excessive is eliminated to a great extent.
Generating an ANFIS structure using subtractive clustering as shown in FIG. 2, to form a preliminary model FIS for diagnosing diesel engine faults, wherein the rules are a combination of one per parameter set, thereby reducing the number of rules and the model structure, wherein n is1、n2……n11The numerical value is automatically generated by a program according to the weight value, and the numerical value determines the membership degree in the ANFIS structure, represents the fuzzy interval segmentation degree of the parameters and has important influence on the judgment of the result. In the process of training a program, the value is adjusted through a large number of training steps, but due to the limitation of the number of parameters and the limitation of the storage amount of the program, when the program is actually operated in a limited training step, the number of rules overflows due to the fact that the number of the parameters is too large and the number of iterations is increased, so that a model cannot be generated.
The method searches the clustering center, judges the influence of the parameters in the judgment through the weight value obtained by an analytic hierarchy process, and automatically divides the membership degree by a program, thereby simplifying the generation process of the model.
Further, the generating of the failure diagnosis preliminary model FIS includes inputtingParameter processing step, judgment rule structure Q0And a complete algorithm structure of the result output step. The FIS generated by training is used for the fault diagnosis preliminary model: generating a rule structure Q using training data pairs0Training the parameters, namely training the precondition parameters and conclusion parameters of the generated FIS by using the complete fault data of the known result set F and the input parameter set S in the fault database, wherein the precondition parameters and the conclusion parameters are parameters related to the generation of a preliminary rule and a final integration rule in the generation of a fault diagnosis preliminary model FIS, and the diagnosis precision of the model is improved by the training parameters;
training is carried out by adopting a hybrid learning algorithm, a back propagation algorithm is applied to precondition parameters, a linear least square estimation algorithm is adopted to adjust parameters to conclusion parameters, input signals are transmitted to the fourth layer along the network forward direction in each iteration, the precondition parameters are fixed, and the conclusion parameters are adjusted by adopting the least square estimation algorithm until the model simulation runs to the fifth output layer;
the obtained error signal is propagated reversely along the diagnostic model ANFIS structure, so as to adjust the precondition parameters; in the invention, the training step number is selected as a training end standard, and the training is ended after the training reaches the specified step number to generate a fault diagnosis structure FIS'.
Furthermore, the fault diagnosis structure is a trained diesel engine fault diagnosis ANFIS structure, the structure is used for actual diesel engine fault diagnosis, during diagnosis, diesel engine fault input parameter data S' and training data S are integrated and then are subjected to normalization processing, the processed data are input into a parameter set S ", the input parameter set S" is input into the fault diagnosis ANFIS structure as an input parameter to be subjected to simulation operation, the obtained result is a fault type, and the fault type is expressed through image and numerical value output. Finally, a diesel engine fault diagnosis model integrating various parameters is formed through data processing and model generation, and a judgment result is finally generated.
The method comprises the steps of firstly carrying out normalization processing on input parameters of an algorithm model, then determining the weight of the input parameters by using an analytic hierarchy process, secondly optimizing the central value of an ANFIS subtractive clustering method by using the weight of the input parameters, and finally establishing an ANFIS-based diesel engine multi-information fusion rapid diagnosis model. The model has the characteristics of more input parameters, small algorithm calculation amount, automatic optimization of the model structure, independence of expert experience and the like. The fault diagnosis model guides operation and maintenance personnel to carry out targeted fault maintenance on the diesel engine, and reduces the maintenance cost of the diesel engine.

Claims (10)

1. A diesel engine multi-information fusion diagnosis method based on ANFIS is characterized by comprising the following steps,
step 1: selecting typical faults of the diesel engine as an output result set F, and selecting kinetic and thermodynamic parameters which are convenient to measure in actual operation of the diesel engine as an input parameter set S; fault database U trained by taking known fault parameter set as model1；
Step 2: the output result set F, the input parameter set S and the fault database U of the step 1 are combined1Normalizing the data set;
and step 3: carrying out analytic hierarchy process on the data set subjected to the normalization process in the step 2 to obtain a parameter weight value;
and 4, step 4: combining the parameter weight values in the step 3 with the ANFIS optimized by the subtractive clustering method to generate an initial rule structure Q of the diesel engine fault diagnosis model0；
And 5: the initial rule structure Q of the step 40Obtaining a rule structure Q of a diesel engine fault diagnosis model after training1；
Step 6: rule structure Q of fault diagnosis model of diesel engine1Inputting characteristic parameters in the diesel engine actual operation parameter set S subjected to normalization processing, recording the characteristic parameters as an input parameter set S ', performing fuzzy fitting and rule fitting on the input parameter set S', and generating a diesel engine fault diagnosis model based on ANFIS.
2. The method of claim 1The diesel engine multi-information fusion diagnosis method based on ANFIS is characterized in that the fault type in the step 1 comprises the following steps: normal operation f0Single cylinder misfire f1Exhaust pipe leakage f2Dirt resistance f of air compressor3Air filter clogging f4Poor lubrication f5；
The input parameters include: effective power s1Fuel consumption s2Front vortex exhaust temperature s3Post-vortex exhaust temperature s4Vortex front exhaust pressure s5Front temperature s of intercooler6And the rear temperature s of the intercooler7Front pressure s of intercooler8And intercooler rear pressure s9Inlet pressure s10After cylinder average temperature s11。
3. The ANFIS-based multi-information fusion diagnostic method for diesel engines as claimed in claim 1, wherein the step 2 is to normalize the data set by using the maximum and minimum method for each group Si in the set S
Figure FDA0002638254670000011
Wherein, i is the number of parameters i is 1, …, 11, k is the number of data groups siminK is the mean of the data sequence, 1, 2, … …; simaxIs the variance of the data; setting the normalization range to be 0-1; normalization eliminates the magnitude difference between different types of input parameters.
4. The ANFIS-based multi-information fusion diagnosis method for the diesel engine as claimed in claim 1, wherein the step 3 of performing analytic hierarchy process on the data set specifically comprises the steps of obtaining a difference value between fault data and normal data, establishing a data characteristic square matrix, obtaining a characteristic value and a characteristic vector of the square matrix, comparing the maximum characteristic vector, and performing consistency detection to obtain a parameter weight value.
5. According toThe ANFIS-based multi-information fusion diagnostic method for diesel engines as claimed in claim 4, wherein the analytic hierarchy process is specifically to find the degree of parameter variation, i.e. the difference between the fault parameter and the normal parameter in the known set, and to use the difference to form a parameter variation matrix A, if the ratio of the difference between the factor i and the factor j is
Figure FDA0002638254670000026
Then the ratio of the difference between factor j and factor i is
Figure FDA0002638254670000021
Because the square matrix A ═ aij)n×nThe requirements are met,
(i)aij＞0，(ii)
Figure FDA0002638254670000022
so that the square matrix A is (a)ij)n×nIs a positive and negative matrix;
but the square matrix A ═ aij)n×nIt is not necessarily the case that,
Figure FDA0002638254670000023
the matrix A ═ aij)n×nNot necessarily a consistency matrix;
according to the consistency matrix characteristics:
if A is ═ aij)n×nMaximum eigenvalue λ ofmaxThe corresponding feature vector is w ═ w (w)1，…，wn)TThen, then
Figure FDA0002638254670000024
Wherein w is the parameter weight value;
by analytic hierarchy process basic structure, before obtaining the weighted value, should construct square matrix A and carry out the conformance test, through the conformance proportion CR:
Figure FDA0002638254670000025
when CR is less than 0.10, the matrix is a consistency matrix, otherwise, the matrix is modified;
the numerical value of CI is calculated and obtained according to the square matrix A, the numerical value of RI is obtained by looking up an RI value table of the analytic hierarchy process random consistency, the numerical value of RI is determined by the order n of the square matrix A, the numerical value of RI can also be obtained by calculating the square matrix A, and the numerical value of RI is obtained by adopting a table look-up method in the invention.
And solving the characteristic value and the characteristic vector of the difference value square matrix A, and carrying out analytic hierarchy process to obtain the weight value of the parameter.
6. The ANFIS-based diesel engine multi-information fusion diagnosis method as claimed in claim 1, wherein the diesel engine fault diagnosis model initial rule structure Q in the step 40Specifically, the training steps and the maximum error are set, and a fault database U is used1For the initial structure Q0Training is carried out, and a rule structure Q of a diesel engine fault diagnosis model after training is generated1。
7. The ANFIS-based diesel engine multi-information fusion diagnosis method as claimed in claim 6, wherein the ANFIS is optimized by subtractive clustering, wherein s1、s2Is an input parameter, and under the condition of multi-parameter input, the structure generation conditions are as follows:
the first layer is a blurring layer, and the expression is as follows:
Figure FDA0002638254670000031
wherein A isi、BiRepresenting a fuzzy set;
Figure FDA0002638254670000032
and
Figure FDA0002638254670000033
are respectively s1，s2Selecting the expression of the bell-shaped function as follows:
Figure FDA0002638254670000034
wherein, { aj，bj，cjThe value of the parameter set is continuously updated by feedback in the training stage, and finally a conclusion parameter set in the rule is formed;
the second layer is used for realizing the operation of the fuzzy set in the first layer, the output of the first layer is represented as points in the layer, the algebraic product of signals is output through the calculation of the layer, the output result of each point is represented as a rule applicability, and the expression is as follows:
Qi＝μAi(s1)×μBi(s2)，i＝1，2 (8)
the third layer is to normalize the excitation intensity of each rule, the node of the layer is a fixed node, and the output is the applicability of the rule and all rules; the expression is as follows:
Figure FDA0002638254670000035
all nodes of the fourth layer are self-adaptive to calculate the output of each rule; the expression is as follows:
Figure FDA0002638254670000036
the fifth layer is an output layer which is used for calculating the sum of all transmitted signals as an output signal; the expression is as follows:
Figure FDA0002638254670000037
for advanced parameter modification, a suitable set of parameters is found such that,
Figure FDA0002638254670000038
wherein f is the actual output;
Figure FDA0002638254670000039
and outputting the model.
8. The ANFIS-based multi-information fusion diagnostic method for the diesel engine as claimed in claim 7, wherein the subtractive clustering method is used in the first layer and the second layer, the subtractive clustering is a density algorithm, which is used for finding the center of data, and the density of each data point is firstly calculated to obtain a density index:
Figure FDA00026382546700000310
finding out the data with the maximum density index as a first clustering center, removing the density of the point, and calculating the density indexes of multiple suspicious points;
Figure FDA00026382546700000311
finding the maximum density index, taking the point as a clustering center, and sequentially circulating until the conditions are met:
Figure FDA00026382546700000312
where is a set, small positive number.
9. The ANFIS-based diesel engine fuel of claim 1The information fusion diagnosis method is characterized in that the generation of the failure diagnosis preliminary model FIS comprises the steps of processing input parameters and judging a rule structure Q0And a complete algorithm structure of a result output step, wherein the FIS generated by training refers to: applying training data to rule Structure Q0Training, namely training the generated FIS by using the complete fault data of the known result set F and the input parameter set S, namely training the precondition parameters and the conclusion parameters, and improving the diagnosis precision of the model by using the training parameters;
for the precondition parameters, a back propagation algorithm is applied, for the conclusion parameters, a linear least square estimation algorithm is adopted to adjust the parameters, each iteration firstly transmits the input signals along the network forward direction until the fourth layer, at the moment, the precondition parameters are fixed, and the conclusion parameters are adjusted by the least square estimation algorithm until the model simulation runs to the fifth layer output layer;
the obtained error signal is propagated reversely along the diagnostic model ANFIS structure, so as to adjust the precondition parameters; in this way, the optimal values of the parameters are found, and the training is finished when the training reaches the specified number of steps, generating the fault diagnosis structure FIS'.
10. The ANFIS-based multi-information fusion diagnosis method for the diesel engine is characterized in that during diagnosis, diesel engine fault input parameter data S' and training data S are integrated and then normalized, the processed data are input into a parameter set S ", the input parameter set S" is input into a fault diagnosis ANFIS structure as an input parameter to be subjected to simulation operation, the obtained result is a fault type, and finally a diesel engine fault diagnosis model integrating multiple parameters and generating a judgment result is formed.
CN202010831743.9A 2020-08-18 2020-08-18 Diesel engine multi-information fusion diagnosis method based on ANFIS Active CN112036079B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010831743.9A CN112036079B (en) 2020-08-18 2020-08-18 Diesel engine multi-information fusion diagnosis method based on ANFIS

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010831743.9A CN112036079B (en) 2020-08-18 2020-08-18 Diesel engine multi-information fusion diagnosis method based on ANFIS

Publications (2)

Publication Number Publication Date
CN112036079A true CN112036079A (en) 2020-12-04
CN112036079B CN112036079B (en) 2021-12-07

Family

ID=73577503

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010831743.9A Active CN112036079B (en) 2020-08-18 2020-08-18 Diesel engine multi-information fusion diagnosis method based on ANFIS

Country Status (1)

Country Link
CN (1) CN112036079B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112597658A (en) * 2020-12-28 2021-04-02 哈尔滨工程大学 Multi-model fault diagnosis method for marine diesel engine based on working condition identification

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009013754A1 (en) * 2007-07-24 2009-01-29 Technion Research And Development Foundation Ltd. Chemically sensitive field effect transistors and use thereof in electronic nose devices
CN105740619A (en) * 2016-01-28 2016-07-06 华南理工大学 On-line fault diagnosis method of weighted extreme learning machine sewage treatment on the basis of kernel function
CN108845217A (en) * 2018-08-23 2018-11-20 四川多成电力工程安装有限责任公司 Bushing shell for transformer monitoring method based on FUZZY NETWORK fusion
CN109539886A (en) * 2018-03-25 2019-03-29 哈尔滨工程大学 A kind of stealthy aid decision-making system of submarine and decision-making technique
CN109932184A (en) * 2019-03-20 2019-06-25 杭州电子科技大学 Marine diesel inordinate wear diagnostic method based on union reliability rule-based reasoning
CN109977621A (en) * 2019-04-30 2019-07-05 西南石油大学 A kind of air-conditioning failure prediction method based on deep learning
CN110297178A (en) * 2018-05-11 2019-10-01 宫文峰 Diesel generating set fault diagnosis and detection device and method based on deep learning
CN110441725A (en) * 2018-05-03 2019-11-12 国网安徽省电力有限公司电力科学研究院 Intelligent electric meter method for diagnosing faults and device based on Adaptive Neural-fuzzy Inference
CN111398832A (en) * 2020-03-19 2020-07-10 哈尔滨工程大学 Bus battery SOC prediction method based on ANFIS model

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009013754A1 (en) * 2007-07-24 2009-01-29 Technion Research And Development Foundation Ltd. Chemically sensitive field effect transistors and use thereof in electronic nose devices
CN105740619A (en) * 2016-01-28 2016-07-06 华南理工大学 On-line fault diagnosis method of weighted extreme learning machine sewage treatment on the basis of kernel function
CN109539886A (en) * 2018-03-25 2019-03-29 哈尔滨工程大学 A kind of stealthy aid decision-making system of submarine and decision-making technique
CN110441725A (en) * 2018-05-03 2019-11-12 国网安徽省电力有限公司电力科学研究院 Intelligent electric meter method for diagnosing faults and device based on Adaptive Neural-fuzzy Inference
CN110297178A (en) * 2018-05-11 2019-10-01 宫文峰 Diesel generating set fault diagnosis and detection device and method based on deep learning
CN108845217A (en) * 2018-08-23 2018-11-20 四川多成电力工程安装有限责任公司 Bushing shell for transformer monitoring method based on FUZZY NETWORK fusion
CN109932184A (en) * 2019-03-20 2019-06-25 杭州电子科技大学 Marine diesel inordinate wear diagnostic method based on union reliability rule-based reasoning
CN109977621A (en) * 2019-04-30 2019-07-05 西南石油大学 A kind of air-conditioning failure prediction method based on deep learning
CN111398832A (en) * 2020-03-19 2020-07-10 哈尔滨工程大学 Bus battery SOC prediction method based on ANFIS model

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WANG J ET AL.: "Exploiting Bayesian networks for fault isolation: A diagnostic case study of diesel fuel injection system", 《ISA TRANSACTIONS》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112597658A (en) * 2020-12-28 2021-04-02 哈尔滨工程大学 Multi-model fault diagnosis method for marine diesel engine based on working condition identification
CN112597658B (en) * 2020-12-28 2022-02-18 哈尔滨工程大学 Multi-model fault diagnosis method for marine diesel engine based on working condition identification

Also Published As

Publication number Publication date
CN112036079B (en) 2021-12-07

Similar Documents

Publication Publication Date Title
CN108038300B (en) Optical fiber state evaluation method based on improved membership function combined with neural network
CN109213121B (en) Method for diagnosing clamping cylinder fault of fan braking system
CN108197648A (en) A kind of Fault Diagnosis Method of Hydro-generating Unit and system based on LSTM deep learning models
CN112597658B (en) Multi-model fault diagnosis method for marine diesel engine based on working condition identification
WO2023142424A1 (en) Power financial service risk control method and system based on gru-lstm neural network
CN110119518B (en) Method for diagnosing engine fault cause by adopting neural network model
CN112418406A (en) Wind power tower inclination angle missing data supplementing method based on SSA-LSTM model
CN113157732B (en) Underground scraper fault diagnosis method based on PSO-BP neural network
CN114322199B (en) Digital twinning-based ventilation system autonomous optimization operation regulation and control platform and method
CN102567782A (en) Neural-network-based automobile engine torque estimation method
CN112036079B (en) Diesel engine multi-information fusion diagnosis method based on ANFIS
CN111273125A (en) RST-CNN-based power cable channel fault diagnosis method
CN112596031A (en) Target radar threat degree assessment method based on knowledge graph
CN112700047A (en) Oil quantity loss prediction method based on BP neural network
CN116625686A (en) On-line diagnosis method for bearing faults of aero-engine
Wei Design of Web based expert system of electronic control engine fault diagnosis
CN110457776B (en) Test strategy rapid generation method based on fault decision network
Lai et al. A data-driven decision-making approach for complex product design based on deep learning
CN114189047A (en) False data detection and correction method for active power distribution network state estimation
CN114169718A (en) Method for improving reliability of wind turbine generator based on state evaluation of wind turbine generator
CN112562312B (en) GraphSAGE traffic network data prediction method based on fusion features
CN111999651B (en) Power performance test method of lithium ion battery for pure electric automobile
CN113359435A (en) Correction method for dynamic working condition data of thermal power generating unit
CN112149355A (en) Soft measurement method based on semi-supervised dynamic feedback stack noise reduction self-encoder model
WO2024045377A1 (en) Deep-learning-based method for predicting residual life of aero-engine

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant